Cuda registration

.gitignore

@@ -1,4 +1,7 @@
 build
+buildcuda
+buildcudadev
+compile
 
 # Dependency folders
 depends/*/
@@ -123,6 +126,11 @@ ClientBin/
 *.pfx
 *.publishsettings
 
+# Nsight Nvidia Eclipse
+.cproject
+.project
+.settings/
+
 # RIA/Silverlight projects
 Generated_Code/
 

CMakeLists.txt

@@ -1,3 +1,4 @@
+#CMAKE_MINIMUM_REQUIRED(VERSION 3.28.3)
 CMAKE_MINIMUM_REQUIRED(VERSION 2.8.12.1)
 
 SET(PROJECT_VER_MAJOR 0)
@@ -45,7 +46,6 @@ OPTION(ENABLE_PROFILING "Collect profiling stats (memory consuming)" OFF)
 IF(ENABLE_PROFILING)
   SET(LIBFREENECT2_WITH_PROFILING 1)
 ENDIF()
-
 IF(MSVC)
   # suppress several "possible loss of data" warnings, and
   # "zero-length array in struct" from libusb.h
@@ -266,6 +266,7 @@ ENDIF()
 SET(HAVE_OpenGL disabled)
 IF(ENABLE_OPENGL)
   FIND_PACKAGE(GLFW3)
+  set(OpenGL_GL_PREFERENCE "GLVND")
   FIND_PACKAGE(OpenGL)
   SET(HAVE_OpenGL no)
   IF(GLFW3_FOUND AND OPENGL_FOUND)
@@ -277,7 +278,8 @@ IF(ENABLE_OPENGL)
     LIST(APPEND LIBFREENECT2_DLLS ${GLFW3_DLL})
     LIST(APPEND LIBRARIES
       ${GLFW3_LIBRARIES}
-      ${OPENGL_gl_LIBRARY}
+      ${OPENGL_GL_LIBRARY}
+      ${OPENGL_LIBRARIES}
     )
     LIST(APPEND SOURCES
       src/flextGL.cpp
@@ -337,53 +339,81 @@ IF(ENABLE_OPENCL)
   ENDIF(OpenCL_FOUND)
 ENDIF(ENABLE_OPENCL)
 
+
 SET(HAVE_CUDA disabled)
 IF(ENABLE_CUDA)
-  FIND_PACKAGE(CUDA)
+  INCLUDE(CheckLanguage)
+  CHECK_LANGUAGE(CUDA)
+#  FIND_PACKAGE(CUDA)
   SET(HAVE_CUDA no)
-  IF(CUDA_FOUND AND MSVC14 AND CUDA_VERSION VERSION_LESS 8.0)
+  #IF(CUDA_FOUND AND MSVC14 AND CUDA_VERSION VERSION_LESS 8.0)
+  IF(CMAKE_CUDA_COMPILER AND MSVC14 AND CUDA_VERSION VERSION_LESS 8.0)
     SET(HAVE_CUDA "no (VS2015 not supported)")
-  ELSEIF(CUDA_FOUND)
+  #ELSEIF(CUDA_FOUND)
+  ELSEIF(CMAKE_CUDA_COMPILER)
     SET(LIBFREENECT2_WITH_CUDA_SUPPORT 1)
     SET(HAVE_CUDA yes)
+    ENABLE_LANGUAGE(CUDA)
+    if(NOT DEFINED CMAKE_CUDA_ARCHITECTURES)
+      set(CMAKE_CUDA_ARCHITECTURES 50 52 75 89)
+    endif()
 
+    #FIND_PACKAGE(CUDAToolkit)
     STRING(REPLACE "\\" "/" NVCUDASAMPLES_ROOT "$ENV{NVCUDASAMPLES_ROOT}")
     STRING(REPLACE "\\" "/" NVCUDASAMPLES8_0_ROOT "$ENV{NVCUDASAMPLES8_0_ROOT}")
-    CUDA_INCLUDE_DIRECTORIES(
+    #CUDA_INCLUDE_DIRECTORIES(
+    INCLUDE_DIRECTORIES(
       "${MY_DIR}/include/"
       "${CUDA_TOOLKIT_ROOT_DIR}/samples/common/inc"
       "${NVCUDASAMPLES_ROOT}/common/inc"
       "${NVCUDASAMPLES8_0_ROOT}/common/inc"
     )
     SET(CUDA_FLAGS -use_fast_math)
     IF(NOT MSVC)
-      SET(CUDA_FLAGS "${CUDA_FLAGS} -Xcompiler -fPIC")
+      SET(CUDA_FLAGS "${CUDA_FLAGS} -Xcompiler -fPIC -D_FORCE_INLINES")
     ENDIF()
     IF(HAVE_CXX11 STREQUAL yes AND CUDA_VERSION VERSION_GREATER 7.0)
       SET(CUDA_FLAGS "${CUDA_FLAGS} -std=c++11")
     ENDIF()
 
     SET(OLD_CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}")
     STRING(REGEX REPLACE "-std=c\\+\\+.." "" CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}")
-    CUDA_COMPILE(CUDA_OBJECTS
+    # Thrust requires exceptions.  If OpenCL from NVidia is used we don't need this flag.
+    STRING(REGEX REPLACE "-fno-exceptions" "" CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}")
+    MESSAGE(STATUS "*************")
+    MESSAGE(STATUS ${CMAKE_CXX_FLAGS})
+    MESSAGE(STATUS ${CUDA_FLAGS})
+    MESSAGE(STATUS ${CUDA_INCLUDE_DIRS})
+    MESSAGE(STATUS ${CUDA_LIBRARIES})
+    MESSAGE(STATUS ${CUDA_OBJECTS})
+    MESSAGE(STATUS "*************")
+    #CUDA_COMPILE(CUDA_OBJECTS
+    LIST(APPEND SOURCES
+      #include/libfreenect2/cuda_registration.h
       src/cuda_depth_packet_processor.cu
       src/cuda_kde_depth_packet_processor.cu
-      OPTIONS ${CUDA_FLAGS}
+      src/cuda_registration.cu
+      #OPTIONS ${CUDA_FLAGS}
     )
     SET(CMAKE_CXX_FLAGS "${OLD_CMAKE_CXX_FLAGS}")
 
     INCLUDE_DIRECTORIES(${CUDA_INCLUDE_DIRS})
 
-    LIST(APPEND SOURCES
-      ${CUDA_OBJECTS}
-    )
+    #LIST(APPEND SOURCES
+    #  ${CUDA_OBJECTS}
+    #)
 
-    LIST(APPEND LIBRARIES
-      ${CUDA_LIBRARIES}
-    )
+    #LIST(APPEND LIBRARIES
+    #  ${CUDA_LIBRARIES}
+    #)
+    #set_target_properties(freenect2 PROPERTIES CUDA_ARCHITECTURES "35;50;75")
+    #SET_PROPERTY(TARGET freenect2 PROPERTY CUDA_ARCHITECTURES OFF)
   ENDIF()
+  #ENDIF(CMAKE_CUDA_COMPILER)
 ENDIF(ENABLE_CUDA)
 
+
+
 # RPATH handling for CUDA 8.0 libOpenCL.so conflict. See #804.
 IF(HAVE_OpenCL STREQUAL yes AND UNIX AND NOT APPLE)
   FILE(GLOB CUDA_ld_so_conf /etc/ld.so.conf.d/cuda*.conf)

doc/README.md

@@ -0,0 +1,47 @@
+# Compile and execute development version
+
+## With CUDA
+* Edit ```CMakeLists.txt``` to assign ```CMAKE_CUDA_ARCHITECTURES``` in accordance with NVidia card.
+
+```
+cd
+mkdir Programs && cd Programs
+git clone https://github.com/NVIDIA/cuda-samples.git
+export CUDA_SAMPLES_HOME=$HOME/Programs/cuda-samples
+
+git clone https://github.com/blackzafiro/libfreenect2.git
+cd ~/Programs/libfreenect2
+mkdir buildcudadev && cd buildcudadev
+cmake .. -DCMAKE_CUDA_FLAGS=-I\ $CUDA_SAMPLES_HOME/Common -DCMAKE_INSTALL_PREFIX=$HOME/freenect2cudadev
+```
+
+Para la Lenovo ideapad Y700 hay que apagar VAAPI:
+```
+cmake .. -DENABLE_VAAPPI=OFF -DCMAKE_CUDA_FLAGS=-I\ $CUDA_SAMPLES_HOME/Common -DCMAKE_INSTALL_PREFIX=$HOME/freenect2cudadev
+```
+
+To find the libraries:
+
+```
+export FREENECT_INSTALL_DIR=$HOME/freenect2cudadev
+export FREENECT_COMPILE_DIR=$HOME/Programs/libfreenect2/buildcudadev
+export LD_LIBRARY_PATH=$FREENECT_INSTALL_DIR/lib:$LD_LIBRARY_PATH
+export PATH=$FREENECT_COMPILE_DIR/bin:$PATH
+```
+
+Test cuda buffer access with:
+
+```
+Protonect cudaccess
+```
+
+
+## Permissions for Kinect
+
+```
+sudo cp ../platform/linux/udev/90-kinect2.rules /etc/udev/rules.d/
+```
+
+Replug Kinect after copying file.
+
+